Documente Academic
Documente Profesional
Documente Cultură
Fig. 1. (a) Hot-rolling mill at The University of Sheffield; (b) Hot-rolling of steel; (c) Model-based integrated
system for microstructure optimisation and control.
In the second stage, the mill carries-out the rolling control. As a result, a “right-first-time” design and
schedule using advanced control strategies such as production of different alloy microstructures can be
Model-based Predictive Control (MPC) to guarantee achieved. Also with this approach, production times
optimal process performance when applying the and costs are reduced compared to those obtained by
deformation profile. Thus, the knowledge integration current methodologies which lack adequate
of both the stock and the rolling mill allows one to capabilities for finding effective rolling schedules.
achieve optimal scheduling and control of the whole As shown in Fig.1(c), the rolling schedule is set-up
process on a repeatable basis for different steel based on microstructure models of the rolled material
microstructures and properties. and not via empirical assumptions.
This paper is organized as follows: Section 2 It is worth noting that only quantitative parameters of
presents the general view of the proposed integrated the microstructure are considered in the optimisation.
system describing the optimisation methodology, the These parameters, for example grain size or
physically-based model of the C-Mn steel alloy and a dislocation density, are critically related to the alloy
description of the self-tuning predictive controller mechanical properties. Indeed, the refinement of the
implemented in the mill. Sections 3 present the grain size provides one of the most important
results from a real-time hot-rolling experiment. strengthening routes in steels; the finer the grain size,
Finally, concluding remarks and further work the higher the resulting yield stress and, as a result,
relating to this overall study are presented in Section increased strength is obtained (Brandon, 1999) . The
4. unique feature of grain size strengthening is that it is
the only strengthening mechanism which also
increases toughness (Honeycombe, 1995). For the
2. THE INTEGRATED SYSTEM FOR above reasons, special attention to the refinement of
OPTIMISATION AND CONTROL the grain size as a main parameter of the
An experimental laboratory-scale hot-rolling mill microstructure has been given in this work.
(Fig. 1), located in the Department of Engineering
Materials at Sheffield University, is used as part of 2.1. The Optimisation of the Rolling Schedule
the research undertaken by IMMPETUS1, which falls
into the investigation of steel and aluminium The scheduling problem is treated here as an
microstructure evolution in hot-rolling. Fig. 1(c) optimisation problem which minimizes the error
depicts the block diagram of the model-based between the desired microstructure and the final
integrated system for microstructure optimisation microstructure in terms of its quantitative
and control of the hot-rolling process. Under this characteristics. The design approach requires three
framework, one first specifies the desired basic components: the stock model, the process or
microstructure and the initial conditions of rolling; physical constraints present both in the stock and in
then, by using thermomechanical and physically- the mill, and the optimality criterion represented by
based models, knowledge integration of both the an “objective” function. The stock model describes
rolled stock and the mill is used (1) to perform a GA- the microstructure evolution of the material during
based optimal search to find the rolling schedule for hot-rolling. Constraints include the limitations of the
a particular microstructure and, (2) to control the forming process and the hot workability of the stock.
rolling process by using model-based predictive Optimality criteria are related to achieving a
particular final microstructure in terms of
1 quantitative elements within “feasible” processing
Institute for Microstructural and Mechanical Process
Engineering: The University of Sheffield. windows for hot deformation. The optimisation
problem is solved by using Genetic Algorithms from austenite to ferrite can be expressed by the
which are known to be a powerful evolutionary following formula (Pickering, 1978):
methodology to solve a variety of optimisation
⎧⎪ −0.5 ⎫
)]⎪⎬(1 − 0.45 ε )
problems, including those in which the objective
function is discontinuous, nondifferentiable,
⎛ dT ⎞
d α = ⎨ a + b⎜ ⎟ [ (
+ c 1 − exp − 0.015d γ
⎪⎩ ⎝ dt ⎠ ⎪⎭
stochastic, or highly nonlinear (Kalyanmoy, 2001).
(5)
The Stock Model. Although this strategy attempts to
be applied to a wide range of steel alloys, consider where dα and dγ are the grain size in the ferrite and
the case of the physically-based model of the C-Mn austenite phase respectively, dT/dt is the cooling rate
alloy, for which the microstructural state is defined during transformation, ε is the strain in austenite, a,
mainly in terms of the grain size and the percentage b, and c are material constants for the C-Mn steel
of recrystallised material (Sellars, 1980). The alloy. Although the above models were determined
statically recrystallised grain size (drex) can be on the basis of very few data, they have proven to be
expressed using the following system equations: robust and form the basis of many such models that
have been since developed.
d rex = 0.5d 00.67 ε −1.0 (ε < ε )
*
The Optimality Criteria. To find the most appropriate
(ε ≥ ε )
(1)
d rex = 1.8 × 10 3 Z −0.15 * design solution, the optimality criterion is formulated
as a series of “objective” functions to be minimized
in order to obtain the specified microstructural
where ε * = 2.8 × 10 −4 d 00.67 Z 0.15 , d0 is the initial
features (Venugopal, et al., 1997). The objective
grain size ε is strain, ε& is strain rate and Z is the functions are lumped together into a single scalar
Zener-Hollomon parameter given by the following optimality criterion (JM) in the following form:
expression:
J M = J 1F + J 2F + L + J NF (6)
Z = ε& exp(312000 / 8.31T ) (2)
where,
where T is the deformation temperature. The volume
fraction recrystallised (χ) is calculated according to
J iF = β i ( x − x d ) 2 i = 1,2, K N (7)
the following system equations:
[
χ = 1 − exp − 0.639(t / t 05 )2 ]
.
(ε < ε c )
(3)
where the superscript F refers to the requirements on
the desired final states of the microstructure. βi is a
χ = 1 − exp[− 0.639(t / t 05 )] (ε ≥ ε c ) weight factor that scales various terms of JM to
express priorities in the overall criterion. In this case
a quadratic cost-function is used when it is desirable
where ε c = 3.92 × 10 −4 d 00.5 Z 0.15 , t is time in that a microstructure feature x achieves a value xd at
seconds and t50 is time for 50% static the termination of the deformation process. In this
recrystallisation. case, the optimisation problem can be defined as
follows:
When recrystallisation is complete, further grain
growth may take place even in the relatively short
Minimize J M = (d rex − d rex d ) 2 (8)
time available between passes. The time dependence
of grain growth may be represented by the following
equation: Subject to
( )
N2 ωmin 10 rpm T(z-1) (1-0.9z-1)2
⎡ −1 2⎤
JC = ∑ ⎢⎣ P( z ) yˆ (t + j ) − w(t + j ⎥⎦ + %rmax 40 P(z-1) (1-0.8z-1)/0.2
j = N1
%rmin 10 Delay 1
[ ]
NU Sampling
∑ (λ ( j )∆u (t + j − 1)2 (14) Pmax 3 kNm
Time
8 ms
j =1
Table 2. Multi-pass rolling schedule; desired ferrite
where ŷ is the predicted output, ∆u is the change of grain size = 10µm
the control action, N1 is the minimum costing
horizon, N2 is the maximum costing horizon, Nu is Pass
Rolling Speed Reduction Temperature
(rpm) (%) (oC)
the control horizon, w represent the future set-point,
λ(j) is the control weighting sequence, and P(z-1) is 1 17 29 1150
inverse model in the model-following context with 2 14 23 1060
P(1) = 1. The minimization of the cost function 3 30 26 975
described in (9) leads to the following projected 4 24 31 893
control increment:
Temp. (oC)
1000
optimal control of the process for different steel
800 microstructures. Further work is being carried-out by
600
300
using Neuro-Fuzzy modelling to link mechanical
properties with microstructure to be included in the
Size (µm)
200
schedule optimal search.
Grain
100
REFERENCES
Frac. Rec. (%)
100
50
Beynon, J. H. and C. M. Sellars (1992). "Modelling
Microstructure and its Effects during Multipass Hot
0 Rolling." ISIJ International 32(3): 359 - 367.
0 10 20 30 40 50 60 70 80
Time (s) Brandon, D. (1999). Microstructural Characterization of
Materials. New York, John Wiley & Sons.
Fig. 3. Computer simulation of the microstructure Clarke, D. W., C. Mohtadi, Tuffs, P. S. (1987).
evolution during the experiment "Generalized Predictive Control - Part I: The Basic
Algorithm." Automatica 23: 137 - 148.
MPC was implemented using a high-speed dedicated Dixit, U. S. and P. M. Dixit (2000). "Application of Fuzzy
microprocessor with 384kb flash memory for the Set Theory in the Scheduling of a Tandem Cold-Rolling
algorithm, and 80kb for program memory. The Mill." Transactions of ASME 122: 494-500.
Gama, M. A. and M. Mahfouf (2006). Speed Control of an
microprocessor also has a dual-port RAM interface
Experimental Hot-Rolling Mill using Generalised
for communication by a high-speed network Predictive Control. International Conference Control
connection offering up to 5Mbit/s data rate. Fig. 4 2006, Glasgow, Scotland.
shows the real-time control of the mill during the Higginson, R. L. and C. M. Sellars (2003). Worked
rolling experiment and the final ferrite-pearlite examples in quantitative metallography, London Maney.
microstructure at room temperature. The speed Honeycombe, R. (1995). Steels. Microstructure and
control and regulation were very good, showing good Properties. London, Butterworth - Heinemann.
robustness and stability under all operating Kalyanmoy, D. (2001). Multiobjetive Optimisation using
conditions. From the rolling torque measurements, it Evolutionary Algorithms, Wiley.
Krauss, G. (2004). Microstructures, Processing, and
could be seen that there was no constraint violation
Properties of Steels. ASM handbook - Metallography and
in any rolling pass. After the experiment, the rolled microstructures, Materials Park, Ohio. 9: 126 - 137.
sample was cut along the longitudinal and tranversal Mahfouf, M. and D. A. Linkens (1998). Generalised
axes, polished, and etched with 5% Nital for Predictive Control (GPC) and Bioengineering. London,
microstructural analysis. Grain size measurements Tylor & Francis.
were carried-out using the mean linear intercept Pickering, F. B. (1978). Physical Metallurgy and the
method (Higginson and Sellars, 2003). The final Design of Steels. London, Applied Science Publishers
ferrite grain size was 8.02 µm (Fig. 4b), very close to LTD.
the desired microstructure and in good agreement Sellars, C. M. (1980). The Physical Metallurgy of Hot
Working. Hot Working and Forming Processes. C. M.
with the developed optimisation mechanism.
Sellars. London, Metals Society: 3-15.
Venugopal, S., E. A. Medina, Malas, J. C., Medeiros, S.
4. CONCLUDING REMARKS
and Frazier, W. G. (1997). "Optimisation of
Microstructure during Deformation Processing using
An integrated model-based strategy to optimise and
Control Theory Principles." Scripta Materialia 36(3):
control the hot-rolling process was proposed. The 347 - 353.
results obtained using the C-Mn steel alloy showed Yang, Y. Y., D. A. Linkens, Talamantes, J. (2003). "Roll
that the knowledge integration of both the rolled Force and Torque Prediction using Neural Network and
stock and the deformation process should allow one Finite Element Modelling." ISIJ International 43(12).
30 18
Rolling Speed (rpm)
20 16
10 14
0 12
-10 10
-20 8
-30 6
0 20 40 60 80 0 20 40 60 80
Time (s) Time (s)
3
Rolling Torque (kNm)
1.5
0 (a)
-1.5
-3
(b)
0 20 40 60 80
Time (s)
Fig. 4. (a) Rolling mill real-time performance; (b) final ferrite-pearlite microstructure